Motivation

Many of the world’s largest and fastest-growing cities–from Karachi (pop. 14 million; 34.6% increase from 2000-2010) to Delhi (22m; 39.4%), Dhaka (15m, 45.2%), Jakarta (10m; 14.8%), Bangkok (8m, 29.1%), Lagos (11m; 48.2%) and Kinshasa (9m, 55.4%)–are located in South Asia and Sub-Saharan Africa with tropical to sub-tropical climates unlike those of most OECD member cities in the global North. As the tropic/subtropic become increasingly urban, industrial and affluent, it is important to consider how energy demand–particularly for thermal comfort–will evolve differently in these places than it has historically across the OECD.

To illustrate the potential for vast differences in energy demand for thermal comfort between cities in the global North and cities in the Tropics/Subtropics, consider Delhi, India. Delhi, with its massive population and very hot climate, is an outlier compared to OECD member cities but typical of South Asia: Peak summer temperatures routinely exceed 40 °C. (104 F.), and intense heatwaves can approach 50 °C. (122 F.). Given the huge temperature differential between outdoor (say 104 F.) and desired indoor air temperature (say 72 F.), and the thermodynamic fact that energy for cooling scales linearly with the temperature differential, cooling a room in Delhi will require twice as much energy as cooling a room in New York where the summer outdoor-indoor temperature differential is typically half that.

In addition to higher temperatures, the summer season is also much longer: in the past year, Delhi had over six times as many cooling-degree days as New York City (again assuming a desired indoor air temperature of 72 °F). Compounded by (a) leaky building envelopes in developing world cities (designed for natural ventilation, not air conditiong), (b) intense heat-island effects (typically less green space), and (c) massive population growth, peak electricity demand in emerging megacities could one day surpass that of their neighbors to the north–not only in aggregate terms because of their population, but also per-capita due to climate, building design and thermodynamics.

Figure 1 provides a map of urbanization rates for cities worldwide with a population greater than 750,000 (UN 2011). Urbanization rates are clearly highest in South Asia and Sub-Saharan Africa. Figure 2 shows the distribution of population growth rates for the same set of cities, grouped by latitude (global North, Tropics and Subtropics). The expected value (e.g. central tendancy) of urbanization rates in the Tropics and Subtropics is clearly distinct from that of the North. Cities in the tropics and subtropics represent a vast new market for energy services such as electrical air conditioning, refrigeration, and consumer appliances.

Population growth, economic growth and the income elasticity of energy services in emerging economies will largely determine global energy demand in the coming decades. As incomes rise, energy demand increases along the extensive margin as households and businesses purchase energy-consuming assets for the first time (Gertler et al. 2013). Likewise, as populations rise, energy demand increases further along the extensive margin as the number of households and busineses increases, producing a multiplicative effect. This will have huge implications on global electricity capacity expansion, technology deployment and environmental impact, as well as regional effects on consumer prices, air quality and service provision. More broadly, it will have implications on the global transition to renewable energy given the limitations of meeting such large and ‘peaky’ demand with non-dispatchable resources such as wind and solar.

Global Energy Service Provision Parity

Over the long run, we must consider the trajectory of economic development in emerging cities as reaching eventual parity with OECD cities. Further, we must think in terms of energy service provision, not just BTU or KWh. Households and businesses do not demand energy per-se, but energy services such as thermal comfort (cooling, dehumidificaiton and heating), food storage (refrigeration), food preperation (cooking, boiling, microwaving), cleaning (washing, drying), productivity (mobile phones and computers), communication (mobile phones and computers), and entertainment (TV and computers). To begin to accurately forecast future energy demand, we must start with a strong understanding of current demand for energy services. This study aims to address that prerequisite by quantifying current demand for thermal comfort in major emerging cities. We focus on thermal comfort as compared to other end-use energy services because it is by far the largest driver of peak demand in the residential and commercial sectors (Segal et al. 1992; Crowley et al. 2003; McNeil and Letschert 2007).

Drivers of Urban Energy Demand

Broadly speaking, economic activity (as measured by GDP) drives annual average energy use, climate drives seasonal variability, and human physiology and local weather drive diurnal variability. This study considers the latter two – climate and weather.

The three most common classes of energy forecast models are regression, time-series and econometric.

Climate Perspective in Energy Demand Forecasts

Daily and seasonal variability in urban electricity demand is caused, to a large degree, by demand for cooling and heating, which is of course driven by meteorological factors (Segal et al.,1992; Thatcher, 2007).

Segal et al. (1992) evaluated the relationship between summer peak energy demand in Israel and a host of pertinent meteorological parameters. Segal demonstrated that a simple linear model with just a few predictors, namely temperature and humidity, performed as well as more complex models with additional predictors.

Crowley et al. (2003) analyzed the role of climate change-driven effects on electricity demand in both residential and commercial sectors, which are considered to be most sensitive to temperature. Using an observed temperature record, Crowley et al. (2003) estimated average energy demand increases of 3.8% in Pennsylvania, New Jersey and Maryland as a result of recent climate change.

Later Thatcher (2007) built a complex demand forecast model with over 50 model parameters, but ultimately only requires daily max/min temperature and relative humidity as input to estimate electricity demand. Thatcher (2007) takes daily max/min temperatures to estimate apparent temperature in 30-minute intervals using sine-exponential technique; then applies a modified linear regression model to predict electricity demand. The model was also applied to estimate how Load Duration Curves (LDC) change as a result of a 1 degree increase in the average temperature in Australian state capital cities.

As exemplified by these studies, there are many excellent energy demand forecast models available in the literature, with special attention given to the effects of weather and climate. However, all of these studies focus on detailed modeling for a particular mature utility and are therefore hard to generalize to other contexts. Our analysis draws upon consensus findings from the literature (e.g. electricity demand in urban areas is largely a function of temperature) to build our own streamlined urban electrciity demand model. The purpose of our model is not precise prediction (we leave that to the electric utilities themselves), but rather a generalized framework that can be applied across multiple, data-sparse cities simultaneously with reasonable accuracy.

Time Series and Econometric Perspectives in Energy Demand Forecasts

Rallapalli and Ghosh (2012) apply a non-staionary time-series model to accurately predict energy demand in all 5 regional power grids of India. Their model out-performs offical forecasts of the Central Electricity Authority of India for both in-sample and out-of-sample prediction.

Jung (1993), Fillipini (1999), Tiwali (2000), Fillipini and Pachauri (2004) and World Bank (2008) apply econometric approaches to estimate the income elasticity of electricity demand in Korea, Switzerland, India, India and India, respectively.

Many different econometric approaches are found in literature, and can be categorized into macro- and micro-level approaches. Macroeconomic approaches use top-down, national/sub-national statistics (e.g. Bose and Shukla 1999), whereas microeconomic approaches use bottom-up household survey data to analyze across different heterogeneous sub-groups (e.g. Tiwari 2000; Pachauri 2004; The World Bank 2008).

Special Considerations in Tropical and Sub-Tropical Cities

Pachauri and Spreng (2004) found that urban households in India change their energy consumption patterns with rising incomes. Households consume more energy per capita and move from less clean-burning energy sources such as biomass and kerosene to higher quality energy sources such as electricity and LPG. Both trends points towards higher peak electricity demand and higher integral energy consumption.

Two excellent studies by the same author (Gupta 2012, 2014) provide insights into the effect of temperature on thermal-comfort seeking behavior in emerging economies. The first, Gupta (2012) adopted a semi-parametric coefficient model that allowed the temperature-electricity relation to vary over time for Delhi. His main findings were that electricity demand is a U-shaped function of temperature and that the cooling demand per unit increase in temperature (MW/°C) is increasing over time. Gupta (2014) applies a similar analysis to 28 Indian states, rather than a single city. Interestingly, Gupta found that the summer electricity demand temperature sensitivity is higher in hotter climate Indian states and conversely, winter electricity demand temperature sensitivity is higher in colder Indian states. He hypothesized that in hotter climates, people have more cooling equipment, and similarly, in colder states people have more heating equipment. He also conjectured that the effect of both hotter and colder weather on electricity demand sensitivity would be more pronounced with higher incomes and GDP/Capita.

This study builds on the work of Gupta (2014) by conducting a global survey of heating and cooling demand, including emerging and developed cities in the US, India, West Africa, and South East Asia.

Objectives

The literature confirms that daily and seasonal deviations from baseload diurnal electricity demand are driven largely by demand for thermal-comfort: ventilation, dehumification, cooling and heating. Real-time demand for thermal comfort is in turn driven by meteorological factors (Segal et al.,1992; Later Thatcher, 2007).

The objective of this study is to answer four key research questions:

  1. What is the current level of demand for heating and cooling services in major emerging cities, as measured by MW/(°C) above a known threshold temperature?
  2. What is the magnitude of seasonal energy consumption for heating and cooling in major emerging cities, as measured by total GWh?
  3. How do per-capita heating and cooling demand compare across cities, as measured by W/(°C x capita)?
  4. How does the share of annual energy consumption used for heating and cooling compare across cities, as measured by a percent of total?

Data

This study combines high resolution (hourly) electricity demand and meteorological data with annual census information for 50+ global cities.

The starting point for identifying major emerging cities was the UN World Urbanization Prospects (2014), subset to the 100 fastest growing cities with more than 2M inhabitants. For comparison, data for US cities was collected from the Federal Energy Regulatory Commission and the US Energy Information Administration.

City Country Load Weather Population
Abidjan Cote d’Ivoire 2010-2013 (a) 2010-2013 (m) 1990-2030 (n)
Accra Ghana 2013-2014 (b) 2010-2013 (m) 1990-2030 (n)
Amman Jordan 2011-2014 (c) 2010-2013 (m) 1990-2030 (n)
Antigua Antigua and Barbuda 2011-2011 (d) 2010-2013 (m) -
Beirut Lebanon 2011-2014 (e) 2010-2013 (m) 1990-2030 (n)
Chandigarh India 2011-2013 (d) 2010-2013 (m) 1990-2030 (n)
Dakar Senegal 2011-2014 (f) 2010-2013 (m) 1990-2030 (n)
Delhi India 2012-2013 (e) 2010-2013 (m) 1990-2030 (n)
Kano Nigeria 2014 (d) 2014 (m) 1990-2030 (n)
Kupang Indonesia 2013 (d) 2013 (m) 2011 (p)
Manila City Philippine 2011-2013 (g) 2010-2013 (m) 1990-2030 (n)
Mbabane Swaziland 2012-2014 (h) 2010-2013 (m) 2010 (p)
Nairobi Kenya 2011-2013 (i) 2010-2013 (m) 1990-2030 (n)
New York City U.S. 2007-2012 (d) 2007-2012 (m) 2007-2012 (q)
Philadelphia U.S. 2009-2011 (r) 2009-2011 (m) 1990-2030 (n)
Singapore Singapore 2013-2013 (j) 2010-2013 (m) 1990-2030 (n)
Tema Ghana 2012-2013 (b) 2010-2013 (m) -
Tokyo Japan 2008-2014 (k) 2010-2013 (m) 1990-2030 (n)
U.S. cities U.S. 2006-2013 (l) 2010-2013 (m) 2001-2013 (o)

(a): Autorite Nationale de Regulation du Secteur de l’Electricite (b): Ghana Grid Company (c): National Electric Power Company (d): Sustainable Engineering Lab (e): Electricite Du Liban (f): Senelec (g): Philipine Electricity Market Corporation (h): Swaziland Electricity Company (i): Kenya Power and Lighting Company (j): Energy Market Authority (k): Tokyo Electric Power Company (l): FERC & EIA (m): National Oceanic and Atmospheric Administration (n): United Nations World Urbanization Porspects - 2014 (o): US Bureau of Economic Analysis (p): UN Data (q): US Census Bureau (r): PJM

Weather Data

High-resolution weather data are indespensible to accurate energy demand forecasts (Segal et al. 1992; Sailor 2001; Crowley et al. 2003; Thatcher 2007). Fortunately, national weather services and climate information centers such as the U.S. National Oceanic and Atmospheric Administration (NOAA), Britain’s Met Office, and India’s Institute for Tropical Meteorology (IITM), collect, curate, analyze and publish meteorological data from thousands of weather stations worlwide.

NOAA offers a wealth of meteorological/climatological information through the NCDC data portal. The data is available on the Online Climate Data Directory website. It is also available via FTP, which is more effecient for batch queries, and is the method used here.

However, handling large meteorological datasets can be unweildy to the uniitiated. To address this issue and make meteorological data more accessible to a wider range of scientists, engineers and practicioners, we developed the weatheR library for the statistical computing language R. The WeatheR libary dramatically simplifies, streamlines and improves the reproducability of our current work. Complete methodological details, step-by-step instructions and example vignettes are available on our github page.

Breifly, weather data was collected as follows:

  • Cities of interest are geo-referenced via the Google Maps API.
  • City coordinates are passed into a nearest-neighbor search algorithim to the find the k-nearest active weather stations.
  • “Best” neighbor is selected from the k-nearest neighbors using multi-objective criteria of geographic proximity and completeness of the meteorological record.
  • “Best” meteorlogical record is chosen, subset, scrubbed and interpolated to yield hourly temperature and humidity observations for each city and period of interest.

Applying the functions of the weatheR library to the cities and years for which load data was collected, hourly temperature time-series were obtained. The below boxplot shows the temperature distributions for these cities.

Demand Data

Hourly electricity demand data was collected from utility companies, system operators or electricity regulatory bodies serving the cities of interest. Data was collected for the past 3 years, if possible.

Data was obtained for 18 non-OECED cities and 21 OECD cities. [Note: The National Capital Territory of Delhi (population 23 million) is served by five geographically-distinct distribution companies and is thus considered as five seperate cities].

Data for Beirut, Lebanon and Amman, Jordan were estimated from national data. The Jordanian utility NEPCO provided monthly ratios of energy consumption for Amman compared to Jordan as a whole. The ratio was approximately 50% for all the months. Amman is the only major city in Jordan.

The Lebanese utility EDL provided ratios of 15% from 8am to 12am, and 22% from 12am to 8am for Beirut. Comparing the two cities, which are less than 150 miles apart as the crow flies, yields surprisingly low per-capita consumption for Beirut (XX/kwh/capita/yr) compared to Amman (XX Kwh/capita/yr). However, the allocation ratios, and thus the per-capita estimates, makes sense when you consider that there are several major cities in Lebanon (Beirut, Zahle, Tripoli and Saida), whereas Jordan has only Amman. Therefore, despite the similarity in electricity demand between Jordan and Lebanon at the national-scale, applying utility-supplied ratios yields significantly higher per-capita demand for Amman compared to Beirut.

Data for Abidjan, Ivory Coast and Dakar, Senegal were estimated from country level loads by the utilities themselves. Monthly ratios of peaks at city-level feeders to peaks at grid-level were applied.

Population Data

Population data was collected from the UN World Urbanization Prospects (2014) and from the US Bureau of Economic Analysis. It is used to normalize energy demand and consumption to a per capita level, as per the Methods section.

Methods

Two main analyses are performed in this contribution for all the cities/year for which data has been collected. The first one consists in estimating the sensitivity of the daily peak demand of electricity to temperatures, and the second one consists in estimating the amount of energy consumed for thermal-comfort seeking (cooling and heating). The below sections detail the key concepts and methodology used to obtain these estimations.

Cooling/Heating Degree Hours (CDH/HDH)

For every city/year combination, Cooling Degree Hours is defined as the sum at every hour of the difference between the recorded temperature and a certain ‘comfortable’ temperature at which no cooling nor heating is required. If this difference is negative, it is taken as 0, as per the below formula:

\[ {CDH} = \sum_{hour=1}^{8760} max\left \{ T_{observed,hour} - T_{threshold}, 0 \right \} \]

In a similar way, Heating Degree Hours is defined as the sum at every hour of the difference between a certain ‘comfortable’ temperature and the observed temperature at that hour. If this difference is negative, it is taken as 0:

\[ {HDH} = \sum_{hour=1}^{8760} max\left \{ T_{threshold} - T_{observed,hour}, 0 \right \} \]

In other words, CDH and HDH represent the number of degrees per year that require respectively cooling and heating to reach the degree of thermal-comford set by the ‘comfortable’ temperatue. This temperature is usually set at 20 °C, however the subsequent sections will derive a threshold temperature that is specific for the city-year considered.

Temperature-Load Curve (TLC) & Threshold Temperature (Tt)

Starting with the pairwise observations of hourly temperature and energy demand (MW) the temperature-load profile of every city-year combination can be obtained. Iterative testing has shown that to obtain a solid profile suitable for a regression fit, temperature-load data must be available for at least the equivalent of 3 months to capture a wide range of temperatures. Therefore, all the collected data were filtered to this criteria and a small number of city-year combinations were omitted as a consequence.

Economic activity is typically lower on weekends compared to weekdays, and electricity demand is commensurately lower. To remove this noise, which is unrelated to temperature, weekends were omitted.

Plotting the energy demand against temperature gives the Temperature-Load Curve (TLC). Depending on the climate and prevalence of heating and cooling appliances, the strength of the temperature-load profile will vary. Cities with distinct heating and cooling seasons (such as New York City, latitude = 40.65°, figure XX) will have a V shaped TLC. The right side of the curve represents the cooling regime: as temperatures increase, demand for electricity increases because of cooling requirements. The cooling regime, and therefore the city’s sensitivity to high temperatures is characterized by a positive coefficient called the cooling coefficient. The left side of the curve represents the heating regime: as temperatures increase, demand for electricity increases because of heating requirements (although in NYC the main heating requirements are met by natural gas/heating oil and not electricity, some heating applicances do use electricity). The city’s sensitivity for cold temperatures is characterized by a negative coefficient called the Heating Coefficient, which tends to be lower in absolute terms than the Cooling Coefficient for the reason stated above. The temperature that separates the cooling and the heating regime is called the Threshold Temperature. It is the transition between heating and cooling regimes. In other words, it is the observed ‘comfortable’ temperature above which perople start cooling and under which people start heating.

By contrast, cities in the tropics and sub-tropics tend to have cooling but no significant heating. For example, Dakar Senegal (latitude = 14.7°) has a distinct cooling seasons as can be seen from the TLC (figure XX).

A third category is cities located in temperate climates with little to no cooling and heating infrastructure and thus no strong relationship between temperature and load. The TLC for these cities will lack a definite Threshold Temperature and the slope of the regression will not be significantly different than zero. As we will see in the result section, several cities have very recently transitionned from non-significant to significant Heating & Cooling, which indicates increased penetration of thermal-comfort appliances (see figures XX).

Cooling/Heating Power Demand

This section aims to quantify the TLC for a set of global cities. Cities are divided in two groups: Heating & Cooling, and Heating or Cooling. For cities for heating or cooling only, a sample linear regression is used to model load as a function of temperature. For cities with heating and cooling, a segmented linear regression is performed where in two linear models (one for cooling, one for heating) are iteratively estimated to minimize the total sum of square root errors. The intercept (Beta0) and Beta coefficients (Beta1) for the two linear models are estimated simultaniously by ordinary least squares given an initial condition of the breakpoint between the two linear regressions. In physical terms, the breakpoint is the threshold temperature (or transition temperature) between heating and cooling seasons. For each city, the initial condition was set equal to the mean temperature observed in that city. Since a large number of city-year combinations are considered in this study, we developed R functions to automate the process. For every city-year combination, the first step of the algorithm is to compute the IQR (difference between the 75th percentile and the 25th percentile) of the temperature distribution. Empirically, we found that for cities with distinct heating and cooling seasons the IQR was higher than 6.8 and that cities with heating or cooling have a temperature IQR less than 6.8.

Figure XX (below) shows the resulting linear regression fit for NYC and Dakar respectively. The coefficients (Heating and/or Cooling), their significance and the x-coordinnate of the break-point (threshold temperature) are extracted and saved for every city at every year. This is done for all cities at daily peak load observations.

This process is repeated for every hour of the day for each city-year combination, that is, fit a linear/segmented regression model to all the weekday/weekend midnight, 1ams, 2ams, etc. resulting in 24 TLCs for each city-year combination. This will be used in the next section to estimate integral energy consumption for heating and cooling.

The coefficients (heating and/or cooling) obtained at every hour for every city-year combination are a measure of demand for thermal-comfort seeking, that is how much power is required to keep residents in city comfortable. Coefficients are expressed in MW/°C and therefore represent the incremental change in load for every 1°C change in threshold temperatures. In subsequent sections, the Cooling Coefficient will be referred to as the Cooling Demand and the Heating Coefficient as the Heating Demand.

Integral Energy Consumption for Heating and Cooling

After estimating the characteristic energy demand for cooling and heating at each hour of the day (0-23) for each year from the TLC, we can reconstruct integral energy consumption for thermal-comfort by multiplying energy demand per °C by the degree hours (equations 1 and 2). Before computing integral energy consumption, a data filter was applied to select city-year combinations containing the equivalent of at least 350 days of hourly observations. 350 days was chosen instead of 365 days to allow a modest tolerance for missing data (a maximum of 2 weeks of equivalent hourly observations).

\[ {(1) Cooling Energy_{city,yr}} = \sum_{hour=1}^{8760} max\left \{ Cooling Demand_{city,yr,hr}*(T_{observed,city,yr,hr} - T_{threshold,city,yr,hr}), 0 \right \} \]

\[ {(2) Heating Energy_{city,yr}} = \sum_{hour=1}^{8760} max\left \{ Heating Demand_{city,yr,hr}*(T_{threshold,city,yr,hr}-T_{observed,city,yr,hr}), 0 \right \} \]

Two important points must be noted concerning the above two equations. First,fFor any city-year-hour combination, if the Cooling Demand or Heating Demand is not significant at the 90% confidence interval, it is set to 0. Second, for cities with only heating or cooling, or no relationship at all, the threshold temperature is taken as the 5th percentile of the temperature distribution to avoid outliers.

Results

The following section presents results and discussion for each of four stated research objectives. Taken together, our results represent the first comparative analysis of electrical heating and cooling demand at the city-scale, including both OECD and non-OECD member cities.

Cooling

A significant cooling signal was detected (90% confidence level) for 35 of 37 cities analyzed. The only cities without a clear cooling signal were Mbabane, Swaziland (elev. 1243m, lat. -26°, long. 31.3°) and Nairobi, Kenya (elev. 1661m , lat -1.3°, long. 36.8°). Both cities are at high elevation, with temperate climates and cool nights, suggesting that electrical cooling is unnecessary much of the year, and thus adoption of capital-intensive A/C is commensurately low. By comparison, Abidjan, Cote d’Ivoire had no detectable cooling signal as recently as 2010, but now has a highly significant (99% CI) cooling demand of approximately 3 Watts/(capita x °C), suggesting very recent uptake of cooling appliances.

In Accra, Ghana, both the effect of temperature on electricity demand, and the significance of that effect are increasing year-on-year. Virtually across the board among non-OECD cities, this holds true: cooling demand, both city-wide and per-capita, are higher now (most recent year data is available) than even just a few years ago (first year data is available). The overall trend is positive, but not strictly monotonic, for Abidjan, Cote d’Ivoire; Accra, Ghana; Chandigarh, India; Dakar, Senegal; and Manila, Philippines; see Table 2 (city-wide) and Table 3 (per-capita).

This suggests significant, latent, unmet demand for indoor thermal comfort services in emerging market cities. As incomes continue to rise, so will penetration of vapor-compression refrigeration window-units (e.g. A/C) and resistive electrical heaters in the near term, and central heating/cooling and reversible heatpumps in the mid- to long-term. Energy demand for cooling, dehumidification and heating will rise accordingly. How high it will ultimately go, is a central question of this research.

As an upper-estimate, we can presuppose that demand for heating and cooling will reach eventual parity with OECD cities of similar climate on a Watt/(°C x capita) basis. That is, a similar level of indoor thermal comfort is expected once a certain level of affluence is attained. Integrating heating/cooling demand over expected HDH/CDH for a given city, yields a reasonable estimate of total annual energy consumption for indoor thermal comfort. This method can be used for historical, current-year or future projections by adjusting the per-capita heating/cooling demand (intensive margin) and heating/cooling-degree hours (extensive margin). Adjustments can be made to reflect change over time along the development spectrum (intensive margin) and the effects of climate change (extensive margin).

As a group, non-OECD cities (n=15) were found to have maxima per-capita cooling demands ranging from 0-12 W/(capita x °C) in all but three cities 1. By comparison, OECD cities (n=18) ranged from 10-140 W/(capita x °C). The median per-capita cooling demand was seven-times higher in OECD compared to non-OECD cities (40.8 versus 5.8 W/(°C x capita), respectively). The univariate distribution of per-capita cooling demands comparing OECD and non-OECD cities yields statistically distinct sets.

The only OECD cities in our study with per-capita cooling demands similar to that of the non-OECD set, were Los Angeles, Sand Diego and Honolulu (at 11, 23 and 19 W/°C x capita, respectively). All three of these locals have very mild, coastal climates with annual average temperatures at a near-perfect 22°C. While climate clearly attenuates or accentuates the intensity of demand for thermal comfort, it cannot explain all the difference. While L.A., San Diego and Honolulu have near perfect climates, they still have per-capita cooling demands 2-4 times higher than emerging economy cities with much more extreme climates, such as Delhi. Based on climate alone, we would expect the opposite.

Within the OECD set, there is substantial variation. Mid-size, relatively sprawling, U.S. cities such as Detroit, MI; Chattanooga, TN; and Omaha, NE; appear to have the highest per-capita demand for cooling, at approximately 100 W/(capita x °C). Population dense Singapore, on the other hand, has a cooling demand of just 13 W/(°C x capita) despite having seven times as many standardized CDH as Detroit, five times as many as Omaha, and three times as many as Chattanooga. In fact, Singapore has the highest number of standardized CDH of any city in our study, with over 70,000 per annum. Result 6 provides a summary table of per-capita peak demand for electrical cooling (and heating) alongside standardized and optimized CDH (HDH) for all cities in our study.

Returning to the three non-OECD cities with substantially higher per-capita cooling demand than their peer-group, we have: Manila, Philippines at 19 W/(capita x °C); Amman, Jordan at 47 W/(capita x °C); and New Delhi (not NCT Delhi as a whole or any of the other districts; just the governmental district) at 40 W/(capita x °C). These three cities may be harbingers of what is to come among peer non-OECD cities analyzed in this study, as well as thousands of other emerging-market cities worldwide, as incomes rise and demand for indoor thermal comfort increases.

For example, New Delhi (NDMC; the seat of government), has a per-capita cooling demand roughly four times that of neighboring parts of the city. This reflects stark differences in the building stock: many large government buildings have been retrofitted for air-conditioning, a departure from traditional open-envelope building design. Cooling demand in neighboring districts of Delhi will likely catch up quickly, as A/C rapidly becomes commonplace in middle-income households and businesses in all quarters of the city.

Amman, Jordan has per-capita cooling demand nine times as high as Beirut, although the cities are less than a 150 miles apart and have similar climates (although Beirut is more temperate given its location on the Mediterranean). Cooling demand in Amman is more similar to that of El Paso, TX, USA at 56 W/(capita x °C), than nearby Beirut.

Altogether, 15 of 37 cities were found to have statistically significant growth (90% CI) in cooling demand over the period of record, including all non-OECD cities except Beirut and Delhi. 11 of 15 of these cities experienced the highest cooling demand on record in the most recent year. This suggests continued, increasing penetration of air conditioning and increasing square-footage of air-conditioned space. The former is likely the driving force in seven non-OECD cities (Abidjan, Cote d’ivoire; Accra, Ghana; Chandigarh, India; Dakar, Senegal; New Delhi, India; Delhi Military Contonement, India; and Manila, Philippines), whereas the latter is more likely in the four OECD cities (New York City, USA; Philadelphia, USA; Tacoma, USA; Indianapolis, USA.). That is to say, business and residences in non-OECD cities are rapidly adopting A/C, whereas non-OECD cities have already reached market saturation for A/C ownership, but continue to build and convert new square-feet into residential/commercial air-conditioned space. In both cases, demand for thermal comfort continues to rise, suggesting a long-tailed distribution. Even after emerging market cities reach asymptotic adoption rates for heating and cooling appliances, energy demand will continue to rise, following a development pattern of increasing residential/commercial air-conditioned real-estate.

From a system-operator perspective, total urban demand for electrical cooling ranges from less than 10 MW/°C in Abidjan, Accra, Antigua, Beirut, Chandigarh, Dakar, Kano and Tema, to over 2000 MW/°C in Tokyo, 300 MW/°C in NYC and Detroit, and ~200 MW/°C in Philadelphia and Manila. Of course, population and economy explain much of that difference, but large disparities remain even on a per-capita basis, as noted above and illustrated in Tables 3 and 4.

Heating

A significant heating signal was detected (90% confidence level) for 22 of 37 cities analyzed: As temperatures decrease below an empirically-derived threshold temperature (unique to each city), electrical demand increases. That is, an inverse-linear relationship is observed between temperature and electricity demand below a the threshold temperature, typically 15-25°C, depending on the city. Cities with no significant heating signal fall into two categories: (1) tropical, coastal or otherwise mild climates with little need for heating, and (2) cities that do require heating for indoor thermal comfort during parts of the year, but have yet to reach significant penetration rates of heating appliance ownership. Category two are of particular interest because they will undoubtedly change significantly over the next several years and decades as incomes rise, the cost of heating appliances come down, and western living standards are sought.

Many non-OECD cities, which are of particular interest in this study, fall into one of the two aforementioned categories. Non-OECD cities that do indeed have significant heating signals include Amman at 28.3 W/(capita x °C), Beirut at 0.3, Chandigarh at 2.1, Delhi W-SW-S districts at 1, Delhi Military Contonement at 7, Delhi NW-N districts at 0.8, Mbabane at 3.6 and Nairobi at 0.4 W/(capita x °C). Interestingly, in Chandigarh and Nairobi, per-capita heating demand were not significantly different from zero as recently as 2011, but have since become significant and increased year-on-year in each of the past three years. We expect this trend to continue (increasing demand, but not strictly monotonic) for years to come as more households and businesses adopt electrical space heating.

Result 1: Urban peak demand for electrical heating and cooling in MW/(∆T)

Urban peak demand for electrical heating and cooling

Footnote: [1] 0-values represent beta coefficients that we feel to reject as different from zero at the 90% confidence level. [2]: maximum = green dot; minimum = red dot; latest value = blue dot.

Result 2: Per-capita peak demand for electrical heating and cooling in W/(∆T x capita)

Per-capita peak demand for electrical heating and cooling

** Result 3** Penetration of heating and cooling appliances

Result 4: Per-capita integral energy consumption for heating & cooling, and normalized energy consumption for heating and cooling.

Note: The following cities/years were lost as a consequence of the 350-days filter: Abidjan-2010, Accra-2013, Beirut-2011, Beirut-2012, Beirut-2013, Beirut-2014, Chandigarh-2011, Chandigarh-2013, Kano-2014, Nairobi-2011, Nairobi-2012, New York City-2007, New York City-2008, New York City-2009, Sacramento-2006, San Diego-2012, San Diego-2013, Tacoma-2006, Tema-2014.

Result 5: Total per-capita electricity consumption

Result 6: Summary Table of Per-Capita Peak Demand and Integral Energy Consumption for Thermal Comfort

City Optimized HDH Standard HDH Heating Demand [W/(°C x Capita)] Heating Energy [kWh/(capita x yr)] Threshold Temperature Optimized CDH Standard CDH Cooling Demand [W/(°C x capita)] Cooling Energy [kWh/(capita x yr)]
Abidjan 217 0 0 0 23.38 31781 59398 1.67 53
Accra 281 0 0 0 24.00 25264 58895 2.83 143.56
Amman 26398 37056 -28.3 469.68 17.62 32198 22059 45.72 1256.62
Antigua 334 0 0 0 23.40 26920 57246 6.74 187.98
Beirut - - -0.3 - 21.84 - - 4.45 -
Chandigarh 25318 19994 -2.12 36.06 20.05 43148 51879 9.6 477.99
Chattanooga 44455 52451 -45.41 1643.63 18.31 27527 20691 143.33 3845.44
City of North Little Rock 58320 48744 -1.15 104.45 21.90 20233 27351 23.83 578.52
Colorado Springs 76729 95634 -6.76 431.94 17.12 15247 8922 35.09 543.89
Dakar 295 1269 0 0 18.75 51995 42031 2.23 145.28
Delhi 18158 15075 0 18.78 21.13 50783 57648 5.04 370.25
Delhi - BRPL 19673 15075 -0.99 31.46 21.66 47612 57648 10.49 569.19
Delhi - BYPL 17367 15075 0 17.66 20.84 52545 57648 6.22 433.85
Delhi - MES 31658 15075 -7.05 97.31 25.20 28687 57626 9.19 516.79
Delhi - NDMC 21501 15075 0 93.48 22.27 44208 57648 39.61 1578.51
Delhi - NDPL 20765 15075 -0.77 21.12 21.96 46071 57640 6.4 437.62
Detroit 81195 91008 -7.8 614.97 18.46 13381 9670 92.76 1232.03
El Paso 34754 39188 -2.15 144.17 18.91 36697 31530 56.64 1596.69
Eugene 31847 83768 -31.15 978.3 12.25 21269 5259 37.96 430.68
Honolulu 492 267 0 0 20.68 37868 43890 11.9 571.49
Indianapolis 56101 80296 -21.69 1188.17 15.53 27676 12658 73.65 1700.36
Kano - - 0 - 14.63 - - 0.5 -
Kansas City 62037 71636 -1.59 107.06 18.21 25577 19460 9.76 219.06
Los Angeles 4909 29381 0 2.75 11.78 58907 11077 8.58 240.25
Manila 274 0 0 0 25.24 27714 72833 17.33 313.36
Mbabane 703 16342 -3.64 8.48 13.00 67343 17627 0 4.09
Nairobi 297 15401 -0.38 0.25 14.33 49229 11833 0 0
New York City 41391 68932 -4.61 174.36 15.15 28212 11268 37.23 845.77
Omaha 89717 93035 -11.81 1031.57 19.47 15645 14304 110.74 2062.6
Philadelphia 81540 69563 -8.84 532.89 21.98 10334 15696 39.17 708.44
Sacramento 55805 53138 -3.64 150.75 20.31 13601 14437 41 516
San Diego - - 0 - 12.16 - - 19.59 -
Singapore 369 0 0 0 25.00 26849 70280 13.34 246.35
Springfield 69646 70495 -7.48 726.41 19.85 18281 17841 95.94 2037.35
Tema - - 0 - 24.00 - - 7.92 -
Tokyo 38576 46980 -19.47 685.93 18.29 24925 18409 56.4 864.85
Tacoma 45238 89358 - - 13.92 11747 2813 - -

Footnote: Optimized HDH/CDH are calculated from the drived city-specific Threshold Temperature. Standard HDH/CDH are calculated assuming a 20°C baseline Threshold Temperature

Result 7: Urban Peak Load Analysis

City Date Hour Temperature [°C] Peak Demand [MW] Peak Cooling Demand [MW] % Demand for Cooling
Abidjan 2013-03-12 21 28.20 635.31 77.17 12.15
Accra 2014-01-16 21 28.50 541.38 39.97 7.38
Amman 2014-08-26 15 31.00 1506.60 697.03 46.27
Antigua 2011-06-29 12 29.00 50.75 3.38 6.66
Beirut 2014-07-24 1 26.00 553.23 59.40 10.74
Chandigarh 2013-06-06 14 28.27 351.96 87.49 24.86
Chattanooga 2013-07-17 17 32.80 1114.97 658.42 59.05
City of North Little Rock 2013-06-27 17 35.00 249.00 139.03 55.84
Colorado Springs 2013-06-27 16 33.30 883.00 334.43 37.87
Dakar 2014-10-28 22 28.00 507.22 80.49 15.87
Delhi 2012-07-05 15 34.00 5359.87 1290.34 24.07
Delhi - BRPL 2012-07-02 16 36.50 2310.70 920.72 39.85
Delhi - BYPL 2012-07-05 15 34.00 1268.51 359.80 28.36
Delhi - MES 2012-07-04 13 34.00 44.61 15.89 35.62
Delhi - NDMC 2012-07-04 15 32.00 351.63 110.14 31.32
Delhi - NDPL 2012-06-22 16 36.50 1481.06 349.05 23.57
Detroit 2008-07-16 16 30.00 11011.00 3956.75 35.93
El Paso 2013-06-27 16 33.30 1750.00 822.70 47.01
Eugene 2013-07-01 17 25.60 366.00 190.87 52.15
Eugene 2013-07-01 18 26.95 366.00 210.46 57.50
Honolulu 2013-10-28 20 29.40 1161.64 111.79 9.62
Indianapolis 2008-09-02 15 29.20 2858.00 1494.34 52.29
Indianapolis 2008-09-02 16 31.70 2858.00 1766.04 61.79
Kano 2014-08-29 2 27.14 267.00 21.86 8.19
Kansas City 2013-08-29 17 35.60 454.00 237.87 52.39
Los Angeles 2013-09-05 17 32.20 5862.00 2049.16 34.96
Manila 2013-05-07 13 31.37 8237.00 1508.93 18.32
Mbabane 2013-07-12 10 19.00 21.83 0.00 0.00
Nairobi 2013-11-14 20 27.00 724.39 0.00 0.00
New York City 2012-07-18 14 30.10 11111.50 5042.87 45.38
Omaha 2013-08-29 17 33.90 2351.00 1070.13 45.52
Philadelphia 2011-07-22 17 37.80 4703.73 3506.83 74.55
Sacramento 2013-07-03 18 33.30 3014.00 1492.99 49.54
San Diego 2013-08-30 15 25.60 4604.00 711.05 15.44
Singapore 2013-06-25 14 27.00 6804.03 144.34 2.12
Springfield 2013-06-27 16 32.80 721.00 339.25 47.05
Tacoma 2013-03-01 19 14.20 640.00 4.24 0.66
Tema 2014-02-14 20 28.00 208.83 5.10 2.44
Tokyo 2014-08-05 14 29.40 49800.00 22551.63 45.28

Appendix

Delhi DISCOM to District Mapping

DISCOM Districts
NDMC New Delhi
MES Military
BRPL W-SW-S
BYPL NE-E-Central
NDPL NW-N

References

[1] Segal M., Shafer H., Mandel M.. Alpert P., & Balmor V. (1992). Climatic-related evaluations of the summer peak-hours’ electrical load in Israel. Journal of Applied Meteorology, 31, 1492-1498.
[2] Christian Crowley., & Frederick L. Joutz. (2003). Hourly electricity loads: temperature elasticities and climate change. 23rd US Association of Energy Economics North American Conference.
[3] Marcus J. Thatcher. (2007). Modeling changes to electricity demand load duration curves as a consequence of predicted climate change for Australia. Energy, 32, 1647-1659.
[4] Bose, R.K., & M. Shukla. (1999). Elasticities of electricity demand in India. Energy Policy, 27(3), 137-146.
[5] Tiwari, P. (2000). Architectural, demographic, and economic causes of electricity consumption in Bombay. Journal of Policy Modelling 22(1), 81–98.
[6] Shonali Pachauri. (2004). An analysis of cross-sectional variations in total household energy requirements in India using micro survey data. Energy Policy, 32, 1723-1735. [7] The World Bank. (2008). Residential Consumption of Electricity in India: Documentation of Data and Methodology.
[8] Pachauri, S. & Spreng, D. (2004) Energy Use and Energy Access in Relation to Poverty. Economic and Political Weekly, Vol. 39, No. 3 (Jan. 17-23, 2004), pp. 271-278
[9] Gupta, (2012). Global Warming and Electricity Demand in the Rapidly Growing City of Delhi: A Semi-Parametric Coefficient Approach, Energy
[10] Gupta, (2014). The Effect Of Development on the Climate Sensitivity of Electricity in India, Discussion Papers in Economics, Indian Statistical Institute, Delhi, India
[11] David J. Sailor., J, Ricardo Munoz. (1997). Sensitivity of electricity and natural gas consumption to climate in the USA—methodology and results for eight states. Energy, 22(10), 987-998.
[12] Joshua D. Rhodes., Wesley J. Cole., Charles R. Upshaw., Thomas F. Edgar., & Michael E. Webber. Clustering analysis of residential electricity demand profiles. Applied Energy, 135, 461-471.
[13] Heck, S. & Rogers, M. (2014). Resource Revolution (1st Edition). New York: Houghton Mifflin Harcourt
[14] Smit,Remes, Manyika, Roxburgh, Restrepo (2011). Urban world: Mapping the economic power of cities. McKinsey Global Institue.
[15] IEA World Energy Outlook 2014 Factsheet
[16] Austin Energy. (2006). Residential Electricity burden: an investigation of American community survey data.
[17] David J. Sailor., J, Ricardo Munoz. (1997). Sensitivity of electricity and natural gas consumption to climate in the USA—methodology and results for eight states. Energy, 22(10), 987-998.
[18] Energy Information Administration. Electricity Power Monthly. http://www.eia.gov/ electricity/monthly/index.cfm
[19] Huang, Y. J., Ritschard, R., Bull, J., Chang, L. (1986), Climatic indicators for estimating residential heating and cooling loads. Lawrence Berkeley Laboratory, Report LBL-21 I01, Berkeley, CA.
[20] Joshua D. Rhodes., Wesley J. Cole., Charles R. Upshaw., Thomas F. Edgar., & Michael E. Webber. Clustering analysis of residential electricity demand profiles. Applied Energy, 135, 461-471.
[21] Li, X., Sailor, D.J.,(1995). Electricity use sensitivity to climate and climate change. World Resource, 7(3), 334-346.
[22] Massimo Filippini., & Shonali Pachauri. (2004). Elasticities of electricity demand in urban Indian households. Energy Policy, 32, 429-436.
[23] Sailor DJ. (2001). Relating residential and commercial sector electricity loads to climate —evaluating state level sensitivities and vulnerabilities. Energy, 26, 645-657.
[24] Scott, M.J., Wrech, L.E., Hadley, D.L. (1994). Effects of Climate Change on Commercial Building Energy Demand. Energy Sources, 16, 317-332.
[25] S. Mirasgedis., Y. Sarafidis., E. Georgopoulou., D.P. Lalas., M. Moschovits., F. Karagiannis., D. Papakonstantinou. (2006). Models for mid-term electricity demand forecasting incorporating weather influences. Energy, 31: 208–227.
[26] U.S. Census Bureau. (1995-present). American Community Survey. http://www.census.gov/acs/
[27] U.S. Census Bureau. Working at home is on the rise. http://www.census.gov/newsroom/releases/pdf/home_based_workers_us_infographic.pdf


  1. Manila, Philippines at 19 W/(capita x °C); Amman, Jordan at 47 W/(capita x °C); and New Delhi (not to be confused with Old Delhi nor NCT Delhi as a whole; only the relatively new governmental district) at 40 W/(capita x °C).